Skip to main content
Summarizing Agent processes large tool outputs and condenses them into focused, security-relevant summaries. It runs automatically when output exceeds a token threshold, ensuring that downstream agents work with concise, meaningful data rather than raw volume.

​
What it does

  • Automatic summarization: triggers when tool output exceeds 10,000 tokens, condensing it before passing it to the next agent
  • Goal-directed condensation: surfaces only security-relevant content based on the task objective, discarding noise
  • Large output handling: pre-truncates extremely large outputs (over 900,000 tokens) before summarization to stay within processing limits
  • Graceful fallback: returns a truncated version of the original output if summarization fails, ensuring no data is lost

​
How it fits in

Summarizing Agent runs transparently in the background whenever a tool produces output too large for efficient processing. It is not invoked directly: it operates as an internal utility that keeps the agent pipeline efficient when working with large scan outputs, lengthy web pages, or verbose tool responses.